GIST Scholar
Not a member yet
30271 research outputs found
Sort by
Enhanced Writability of 4P4N CFET SRAM Cell With Transmission Gates
No full textThe conventional complementary field-effect transistor (CFET) static random access memory (SRAM) cell with a 4P2N configuration features two access pMOSFETs, leaving spaces for two nMOSFETs above the access transistors intentionally unused, resulting in suboptimal utilization of available space. To address this, we introduce a split-gate process enabling the fabrication of transmission gates. We propose a novel 4P4N SRAM structure with backside contacts (BCs) that significantly enhances writability while maintaining readability. Compared with 4P2N and 4N2P with BCs, 4P4N has higher read delay and energy consumption but shows 54.7% improvement in write performance over 4P2N and 48.1% over 4N2P. In the case of fast NMOS/slow PMOS (FNSP) under V-T variation at process corners, 4P4N enhances read static noise margin (RSNM) while maintaining strong write static noise margin (WSNM).FALSEsciescopu
Non-targeted Analysis of Leachates from Mulch film Microplastics under UV Exposure
No full textMicroplastics are increasingly accumulating in soil environments, with mulch films identified as a major source. However, limited information is available on the chemical compounds leached from these mulch film- derived microplastics into soil. In this study, compounds leached from PE mulch films (black and white) and biodegradable mulch films (PLA+PBAT) were analyzed under UV exposure, using water and ethanol as leaching solvents. Even though biodegradable mulch films are considered alternatives to conventional PE films, they exhibited a greater number of chemical features in their leachates. To identify the leached compounds, library matching and in-silico fragmentation integrated with a plastic-specific database were applied. Additionally, Molecular networking analysis was employed to visualize structurally related chemical features and predict potential transformation products formed under environmental conditions. As a result, 8 compounds were tentatively identified from the PE black film and 8 compounds from the PE white film, while 16 compounds were identified from the biodegradable film. UV exposure increased the chemical complexity of leachates from both PE and biodegradable films. Notably, 9-Octadecenamide, N,N-bis(2-hydroxyethyl)-, (9Z)-, a substance classified as hazardous, was detected only after UV exposure. These findings highlight the importance of including UV exposure conditions when assessing the chemical risks of plastic mulch films in soil environments. Furthermore, they underscore the need for systematic characterization of mulch film leachates to better understand their potential environmental impacts.MasterI. Introduction 1
II. Materials and methods 4
2.1 Chemicals and methods 4
2.2 Preparation of microplastics from mulch films 4
2.3 UV exposure of microplastics 4
2.4 Leaching experiment 4
2.5 Scanning Electron Microscopy 5
2.6 Instrumental analysis (UPLC-QTOF-MS) 5
2.7 Data processing and compound identification 5
2.8 Molecular spectrum networking 7
2.9 Investigation of identified compounds 8
III. Results and discussion 9
3.1 Surface morphological changes of mulch film microplastics before and after UV exposure 9
3.2 Comparison of leaching features 10
3.3 Identification of leachates from mulch film microplastics 10
3.4 Survey of identified compounds 15
3.5 Quantitative analysis of compounds in mulch film microplastic leachate before and after UV exposure 19
3.6 Identification of transformation product candidates 27
IV. Conclusion 28
V. References 29
IV. Acknowledgement 3
Deep Representation Learning of Electronic Health Records for Cardiovascular Disease Patients
No full textElectronic Health Records (EHRs) contain rich longitudinal clinical data that are vital for the advancement of precision medicine; however, their inherent complexity presents significant analytical challenges. This thesis investigates deep representation learning methodologies for EHR data, with a particular focus on cardiovascular risk modeling. Specifically, it addresses two distinct clinical applications: (1) individualized antiplatelet therapy selection for patients with type 2 diabetes mellitus (T2DM), and (2) temporal risk stratification for repeat percutaneous coronary intervention (PCI) among patients who have undergone initial PCI.
In the first application, phenomapping techniques are employed to guide the selection between cilostazol and aspirin in T2DM patients. Conventional phenomapping methods relying on raw clinical features are insufficient to capture complex patient heterogeneity. To address this, a graph-based phenomapping framework is proposed using GraphSAGE, an inductive graph neural network, to model higher-order clinical relationships and generate expressive patient embeddings. These representations enable improved patient stratification, and survival analyses—including log-rank tests—demonstrate that the learned embeddings reliably identify subgroups with differential responses to cilostazol and aspirin.
The second application introduces DA-RNN-Surv, a dual-attention recurrent neural network (RNN) designed to model longitudinal EHR trajectories for survival prediction. This approach addresses key limitations of traditional models such as Cox and DeepSurv, which rely solely on static baseline covariates. By incorporating temporal and feature-level attention, DA-RNN-Surv captures dynamic clinical risk and provides interpretable predictions. Empirical evaluations demonstrate superior predictive accuracy and clinically meaningful attention patterns, particularly for long-term repeat PCI prediction.
In conclusion, this thesis demonstrates the utility of deep representation learning in enhancing cardiovascular risk modeling from EHR data. The proposed graph-based and temporal attention-based embedding strategies improve both treatment personalization and survival prediction, offering clinically actionable insights to support precision cardiovascular care.|전자건강기록(EHR)은 정밀의료를 실현하는 데 필수적인 종단적 임상 데이터를 포함하고 있으나, 그 복잡성과 비정형성으로 인해 기존의 분석 기법으로는 효과적인 활용에 한계가 있다. 본 논문은 심혈관 질환 위험 예측을 위한 EHR 데이터 기반 심층 표현 학습(deep representation learning) 기법을 탐구하며, 다음의 두 가지 주요 임상 응용 사례를 중심으로 연구를 수행하였다. 첫째는 2형 당뇨병 환자를 위한 항혈소판제 맞춤 치료 전략 수립이며, 둘째는 관상동맥중재술(PCI)을 받은 환자의 재시술 위험을 시계열 기반으로 예측하는 전략 개발이다.
첫 번째 연구에서는 2형 당뇨병 환자에게 최적의 항혈소판제를 선택하기 위한 수단으로서 표현형 매핑(phenomapping) 기법을 적용하였다. 기존 임상 변수에 기반한 표현형 매핑 방식은 환자 간의 복잡한 이질성을 효과적으로 반영하지 못하는 한계가 있다. 이를 극복하기 위해 본 연구에서는 GraphSAGE 기반의 귀납적 그래프 신경망을 활용하여 고차원의 임상 관계를 모델링하고 환자 간 표현 임베딩을 생성하는 새로운 그래프 기반 표현형 매핑 프레임워크를 제안하였다. 그래프 기반 임베딩을 활용한 결과, cilostazol 또는 aspirin 치료 혜택이 큰 환자군을 보다 정확하고 효과적으로 구분할 수 있었으며, 생존 분석과 log-rank 검정을 통해 그 유효성을 검증하였다.
두 번째 연구에서는 기존 생존 분석 모델이 정적인 기준 시점 정보에 의존하여 시술 이후의 임상 궤적을 반영하지 못한다는 한계를 극복하고자 하였다. 이를 위해 반복 신경망(RNN) 기반의 생존 예측 모델인 DA-RNN-Surv를 제안하였다. 본 모델은 다시점 EHR 데이터를 입력으로 받아 시계열 정보를 임베딩하며, 시간 및 변수 수준의 attention 메커니즘을 통해 임상적으로 해석 가능한 예측 결과를 제공한다. 실제 PCI 환자 데이터를 기반으로 한 실험에서, DA-RNN-Surv는 Cox 및 DeepSurv보다 우수한 예측 성능을 보였고, 특히 장기 추적 관찰 기간에서 더 높은 정확도와 임상적 타당성을 입증하였다.
결론적으로, 본 연구는 심층 표현 학습이 EHR 기반 심혈관 위험 예측 모델의 성능과 해석 가능성을 모두 향상시킬 수 있음을 보여준다. 제안된 그래프 기반 및 시계열 attention 기반 임베딩 전략은 치료 개인화 및 생존 예측의 정확성을 높여, 정밀의료 실현을 위한 실질적인 기반을 제공한다.Master1 Introduction 1
1.1 Electronic Health Records 1
1.2 Antiplatelet Therapy in Type 2 Diabetes 1
1.3 Percutaneous Coronary Intervention 2
1.4 Research Objectives 3
2 Related Works 4
2.1 Survival Analysis 4
2.1.1 Kaplan–Meier Survival Analysis 4
2.1.2 Log-Rank Test 5
2.1.3 Cox Proportional Hazards Model 5
2.1.4 DeepSurv 6
2.2 Phenomapping for Personalized Treatment Decisions 7
2.3 Graph Neural Networks and GraphSAGE 8
3 Graph-Based Representation Learning for Phenomapping 9
3.1 Background 9
3.2 Materials and Methods 10
3.2.1 Study Population and Data Collection 10
3.2.2 Data Preprocessing 12
3.2.3 Patient Group Definition and Propensity Score Matching 12
3.2.4 Graph-based Patient Embedding 14
3.2.5 Phenomapping Strategy 15
3.3 Experiments and Results 17
3.3.1 Baseline Characteristics 17
3.3.2 Comparison of Traditional and Graph-Based Phenomapping 18
3.4 Conclusion 21
4 Temporal Representation Learning for Survival Modeling 22
4.1 Background 22
4.2 Materials and Methods 23
4.2.1 Study Population and Data Collection 23
4.2.2 Data Preprocessing 23
4.2.3 Dual-Attention RNN Survival Model 24
4.3 Experiments and Results 27
4.3.1 Experimental Setup 27
4.3.2 Predictive Performance Comparison 28
4.3.3 Interpretation of Visit- and Feature-Level Contributions 29
4.3.4 Clinical Validation of Attention-Based Features 32
4.4 Conclusion 34
5 Discussion 35
Summary 37
References 38
Acknowledgements 4
Rare germline JAK/STAT variants in a Korean cohort amplify innate immune responses to vaccination
No full textThe JAK/STAT signaling pathway regulates cytokine-driven responses. Here we investigated the biological impact of germline single nucleotide variants in JAK kinases and STAT transcription factors in vaccine response. First, we applied a data-mining strategy to RNA-seq datasets from a Korean cohort to uncover JAK/STAT germline variants and analyzed their corresponding interferon transcriptomic responses. Ultra-rare variants associated with heightened interferon transcriptomic responses were identified. AlphaFold 3 predicted conformational alterations in these JAK and STAT variants, focusing on protein-protein interactions and receptor-complex assembly. Co-occurring variants in TYK2 and other interferon signaling regulators that could influence the impact of the JAK and STAT variants were explored. We found that data mining can reveal candidate germline variants with potential roles in innate immunity and assessed how analyzing vaccine-induced gene expression changes can serve as an in vivo method to functionally assess the biological significance of missense mutations in key immune signaling pathways. © 2025 The Author(s)TRUEsciescopu
Backbone Assignments of Human MCM6 NTD1 and Evaluation of Its Interaction with Peptides Originated from BLM Helicase
No full textMCM6 is a core subunit of the eukaryotic MCM2 to 7 helicase essential for DNA replication and often overexpressed in various cancers. We report backbone assignments for the human MCM6 N-terminal domain 1 (NTD1), spanning residues 15-115, and test its binding to BLM peptides MBD-N and MBD-D. The [1H- 15N] HSQC spectrum indicates that the MCM6 NTD1 is well folded. Chemical shift–based analysis supports a compact α/β architecture consistent with AlphaFold and cryo-EM models of the MCM complex. HSQC titrations with both BLM peptides show no significant chemical shift perturbations, indicating no detectable binding under the conditions used. These data suggest that additional regions or oligomeric context are required for stable MCM6–BLM interaction.FALSEkc
Advancement of Urea Elimination Using Pre-Halogenation Processes in UV254/Bromine and UV254/Chlorine Systems
No full textThis study evaluated the efficiency and practical applicability of an advanced oxidation process (AOP) that combines pre-halogenation using bromine or chlorine with ultraviolet (UV) irradiation for the effective removal of urea in ultrapure water (UPW) production systems. The removal of trace-level urea, a refractory nitrogenous compound, is essential to meet the stringent water quality requirements of the semiconductor and pharmaceutical industries—yet conventional water treatment methods are often inadequate. To address this challenge, this study developed a pre-halogenation approach wherein urea was converted into highly UV-reactive N-halogenated intermediates (i.e., chloroureas and bromoureas) via chlorination or bromination, followed by rapid photolytic degradation under UV exposure.
Experimental results demonstrated that the pre-halogenation + UV process significantly outperformed conventional treatment strategies. Bromine-based pre-treatment showed superior removal efficiencies over a broad pH range compared to chlorine-based systems. Complete urea elimination (~100%) was achieved under optimized conditions ([bromine]₀/[urea]₀ = 2–5, pre-contact time = 10–60 min). Even with short pre-bromination times (1–10 min), sufficient accumulation of intermediates allowed for fast and efficient urea degradation. Under low initial urea concentrations (1 μM), effective removal was still achieved within 10 minutes by maintaining appropriate bromine dosing. Moreover, the process exhibited high efficiency with relatively low UV energy input, indicating substantial economic and environmental advantages.
Overall, this study demonstrates that the integration of bromine-based pre-halogenation with UV irradiation is a highly effective and sustainable AOP for urea removal in UPW applications. Future work should focus on optimizing operational parameters and evaluating long-term stability under realistic system conditions to facilitate full-scale implementation.MasterChapter 1. General Introduction 1
1.1 Background 1
1.2. Research objectives 5
Chapter 2. Literature Review 6
2.1. Overview of Urea Elimination Challenges and Research Direction for Ultrapure
Water 6
2.1.1. Ultrapure water: characteristics and its role in advanced manufacturing industries 6
2.1.2. Importance of urea elimination in ultrapure water production 9
2.1.3. Characteristics of urea 11
2.2. Recent Advances and Evaluation of Existing Technologies for Urea Removal 13
2.2.1. Adsorption 15
2.2.2. Membrane separation 16
2.2.3. Hydrolysis 16
2.2.4. Photocatalytic oxidation 17
2.2.5. Electrochemical oxidation (EO) 19
2.2.6. Wet air oxidation (WAO) 20
2.2.7. Biological processes 20
2.2.8. Advanced oxidation processes (AOP) 21
2.3. Urea Removal via AOP-Based UV/Chlorine and UV/Bromine Processes 27
2.3.1. Comparison of the physicochemical properties of chlorine and bromine 27
2.3.2. Urea elimination via UV-based processes 30
2.3.3. Urea elimination via UV/chlorine and UV/bromine process 32
2.3.4. Urea elimination via pre-halogenation + UV process 32
Chapter 3. Material and Methods 34
3.1 Chemicals and Materials 34
3.2 Experimental Methods 36
3.2.1 Experimental method for urea chlorination and bromination treatment 36
3.2.2 Experimental method for simultaneous UV/chlorine and UV/bromine processes 36
3.2.3 Experimental method for pre-halogenation and UV photolysis processes 37
3.3 Analytical Methods 38
3.3.1 Total urea analysis (DAMO and xanthydrol methods) 38
Chapter 4. Results and Discussion 40
4.1 Comparative Evaluation of Urea Removal by Chlorination and Bromination 40
4.1.1 pH-dependent kinetics of urea chlorination 40
4.1.2 pH-dependent kinetics of urea bromination 41
4.2 Comparison of Sequential and Simultaneous UV/Oxidant Processes for Urea
Elimination 44
4.2.1 Comparative evaluation of UV/chlorine and UV/bromine processes under varying pH
and oxidant concentrations. 44
4.2.2 Absorbance monitoring of halogenated product formation 46
4.2.3 Comparative evaluation of sequential and simultaneous UV/oxidant processes for total
urea elimination under varying pH and oxidant conditions 48
4.3 Optimization and Practical Applicability of Pre-bromination Combined with UV
Processes 55
4.3.1 Optimization of pre-bromination time and bromine dose for enhanced urea elimination
at pH 7 55
4.3.2 Urea elimination at trace concentrations via pre-bromination + UV processes 61
4.4 Practical Implication for UPW Production 66
Chapter 5. Conclusions 68
References 70
Acknowledgement 83
Curriculum Vitae 8
Functional Characterization and Heterogeneity Analysis of Ribosomal Proteins in Mouse Preimplantation Embryos
No full textTranslational control is important during the mammalian preimplantation phase, when maternal RNAs and proteins are degraded and de novo synthesis of RNAs and proteins increases. Proteins are synthesized in ribosomes, which are assembled from ~82 ribosomal proteins (RPs). The function of ribosomes varies depending on the resident RPs, suggesting that ribosome heterogeneity can lead to functional specialization. Only a few studies have investigated the function of RPs during preimplantation embryonic development. Here, we performed functional analyses on six RP-encoding genes—Rpl4, Rps9, Rps11, Rpl13a, Rpl19, and Rpl39—in mouse preimplantation embryos. Knockdown (KD) of each of these RP genes, except Rpl39, affected morula-to-blastocyst transition, producing phenotypes that varied somewhat in their details. Rpl4-, Rpl13a-, and Rpl19-KD embryos showed fragmentation and strong arrest of cell proliferation, whereas Rps9- and Rps11-KD embryos showed severe fragmentation with relatively weak arrest of cell proliferation. In the case of Rpl39, single-KD embryos developed normally, but double-KD embryos with its paralog Rpl39-like (Rpl39l) inhibited normal blastocyst development. Protein misfolding signals were also activated in Rpl39-KD and Rpl39l + Rpl39 double-KD embryos, confirming a previous finding that RPL39 and RPL39L are associated with ribosome exit tunnels. Our results suggest the presence of different groups of proteins that require an RPL39-containing ribosome or RPL39/RPL39L-containing ribosome for correct folding in early embryos. Taken together, the results of the present study demonstrate that ribosomal proteins are fundamentally important for normal blastocyst formation and development, but not all ribosomal proteins contribute equally to embryonic development, providing a novel example of ribosome heterogeneity in preimplantation embryos. © 2025 The Author(s). The FASEB Journal published by Wiley Periodicals LLC on behalf of Federation of American Societies for Experimental Biology.TRUEsciescopu
Therapeutic Potential of Novel Antimicrobial Peptide Pap12-6-10: Mechanisms of Antibacterial and Anti-inflammatory Action Against Gram-Negative Sepsis
No full textTo develop novel antibiotics, we engineered 12-mer peptides derived from the N-terminus of papiliocin. Pap12-6-10 emerged as a potent antibacterial agent against multidrug-resistant Gram-negative bacteria, demonstrating a low propensity for resistance development. Pap12-6-10 exerts its antibacterial activity by permeabilizing bacterial membranes through binding to lipopolysaccharide (LPS), inducing oxidative stress that leads to cell death. Pap12-6-10 modulates LPS-induced inflammatory responses by selectively targeting the TLR4 signaling pathways. Structural analysis using NMR, surface plasmon resonance, docking, and molecular dynamics simulations suggested that Pap12-6-10 binds to the hydrophobic pocket of MD-2, thereby preventing the LPS-induced dimerization of the TLR4/MD-2 complex, which is essential for inflammatory signaling during sepsis. In the Escherichia coli K1 and carbapenem-resistant Acinetobacter baumannii-induced sepsis mouse model Pap12-6-10 protected organ damage from septic shock and displayed significant therapeutic effects while maintaining low cytotoxicity. This study highlights its potential as a valuable candidate for treating Gram-negative infections.FALSEsciescopu
Sustainable, highly effective selenate removal using electrochlorination facility–obtained magnesium precipitate: An approach to in situ layered double hydroxide formation
No full textWe proposed a sustainable selenate (Se(VI)) removal method using in situ layered double hydroxide (LDH) formation and calcined magnesium precipitate (CMP) derived from an electrochlorination facility. During kinetic experiments, the optimal removal conditions were identified by adjusting pH and Al dosage, which enabled a notable Se(VI) removal efficiency (127 mg/g-CMP). The initial release of Mg ions from CMP was confirmed to occur via Al-ion hydrolysis. Subsequently, after the pH adjustment process, Se(VI) could be sequestered via outer-sphere complexation within the interlayer space of the brucite-like sheets in LDH. The reusability of precipitated sludges (LDHS and calcined LDHS (LDOS)), which can be secondary wastes, was also evaluated as Se(VI) adsorbents. The results confirmed the high maximum adsorption capacity of LDOS (42.1 mg/g), demonstrating its performance comparable to that of reported adsorbents. This study highlights the potential of the in situ LDH formation method for Se(VI) removal and the possibility of transforming magnesium precipitate into sustainable resources. © 2025 The AuthorsTRUEsciescopu
Au@h-BN Core–Shell Nanostructure as Advanced Shell-Isolated Nanoparticles for In Situ Electrochemical Raman Spectroscopy in Alkaline Environments
Get PDFRecent advancements in in situ electrochemical Raman spectroscopy using shell-isolated nanoparticles have facilitated direct analysis of electrochemical mechanisms. However, shell materials such as SiO2 and Al2O3 commonly adopted for shell-isolated nanoparticle-enhanced Raman spectroscopy are unstable and unreliable in alkaline environments, posing significant obstacles for relevant research in the alkaline environment. While alternative shell materials have been explored, finding suitable replacements for traditional SiO2 shells is still challenging. To address this issue, this study proposes hexagonal boron nitride (h-BN), with atomically ultrathin and insulating properties, as an alternative shell material. Specifically, pinhole-free Au nanoparticles coated by an h-BN shell (Au@h-BN) with a uniform thickness of 1 nm are synthesized through a controlled two-step process. The resulting Au@h-BN exhibits more pronounced Raman scattering and long-term stability under alkaline conditions compared to Au@SiO2. Theoretical simulations support a stronger electromagnetic field distribution around Au@h-BN compared to that around Au@SiO2. In situ Raman studies conducted during electrochemical reactions of Ni and Cu electrodes demonstrate the superior Raman enhancement effect and durability of Au@h-BN compared to Au@SiO2. These results suggest that Au@h-BN holds significant potential for advancing long-term in situ Raman studies in alkaline systems, supporting the development of efficient catalysts for sustainable energy applications.TRUEsciescopu